Multi-core performance vs single-core performance.

[ZPett]

I went to college for computer programming and if there is anything I have learnt from my efforts to make better more optimized programs it is this:

Many cores is not all you need. You need both single core performance and multi core availability to achieve better running programs. One of the reasons I just bought an Intel 9th gen CPU over an AMD was because of their exceptional single-core performance and somewhat good multi-core performance ratio.

What I have learnt in my multi-core programming experiences:

If you write a program to run in parallel and across multiple cores, it always has to branch from a single process called a host process. The host process then generates child processes that run in parallel with each other across the cores. The host process usually needs to manage the returns of the child processes by waiting or watching for specific results from each of the child processes. This means the host process might have to be running loops and constantly cycling through the child processes to check for specific return values. If you don't have good single core performance, there will be a major bottleneck in the multiprocessing program. It got me thinking why you "Intel" haven't made a consumer level motherboard architecture "chip-set" that has 2 CPU sockets where one chip is a fat single core at a high clock rate and the other chip is a dense multi-cored chip at a lower clock rate. It seems the more cores, the more heat so the clocks have to drop. On the single-cored chip at 10 nano-meters you will have more room to spread the transistors out so the juice flowing through isn't so concentrated which would help it dissipate the heat better and enable you to achieve much higher clock speeds. To reduce the latency between the chips, why not try sandwiching the motherboard with a chip directly across from each other on either side? Of course this would require a custom case with a hole behind the motherboard area where the CPU is like the Corsair Air 540. Anyways, I believe this would really help balance the performance of a computer system. The AMD thread-ripper with 32 cores is nice, but why doesn't its performance evenly scale in speed by all the cores that are added? The reason is the single core performance bottleneck that is happening. There is a host process that initiates, controls and finalizes a parallel program operation and if it is running just as fast as the other cores (especially if there are many) it can cause a bottleneck.

Since the nature of a chip is basically to route and process a binary flow of voltage, then a binary tree is how the chip core hierarchy should be structured. Have 1 big and fast core, then 8 that are 1/8 as big and fast, then 64....so on and so forth. They haven't tried a Sierpinski carpet core pattern yet? Give me a break... The billionaires and military's save the good stuff for themselves. That's where this all came from right?

Having 2 chips on the same board at different clocks would most likely need 2 separate memory controllers and 2 separate memory types and speeds which could also require an OS to be rewritten to handle it. I don't know a whole lot about computer architecture and this could just cause a myriad of issues and could be completely ridiculous but I'm just trying to give you guys some new ideas.

[RAJU2529]

thanks for the article , mostly you are research student , go further , wish you all the best.

According to your point , what type of processor would you recommend to the consumer , please mention names of processor models

[ZPett]

Depends what they need to compute because Intel and AMD have chips made for different tasks and there really isn't a chip that you can get the best of both worlds from yet. So the answer is an if, else statement.

I do intense computing and rendering in 3D applications like Blender but I also like to play games occasionally at max settings. I would need different specs in 2 different computers to do both effectively.

There are so many tiers of chips specific for different computing needs and it seems that that margin is increasing. I believe the answer to narrow it is the type of thinking from what I typed above. Why create twice as much junk after 8 years of outdated hardware if you don't need to?